Professional Documents
Culture Documents
Technical Report
Technical Summary
Context
This technical material was developed in October, 2013 by members of the North
American SynchroPhasor Initiative, a collaboration between the North American electric
industry (utilities, grid operators, vendors and consultants), the North American Electric
Reliability Corporation, academics, and the U.S. Department of Energy, to advance and
accelerate the development and use of synchrophasor technology for grid reliability and
efficiency. The material attached was produced for one of a series of NASPI technical
workshops intended to educate and document the stakeholder community on the state of
the art for key synchrophasor technology issues.
In 2009, the Department committed a total of $412 million of funds from the American
Recovery & Reinvestment Act of 2009 to twelve Smart Grid Investment Grants and one
Smart Grid Demonstration Project that implemented and tested synchrophasor
technology using matching private funds. While some of the ARRA funds was spent on
other transmission assets, in aggregate over $328 million of federal and matching private
investment was spent on synchrophasor technology and related communications
networks.
Additionally, DOE has funded significant technical assistance for NASPI and
synchrophasor advancement through the National Laboratories and the National Institute
for Standards & Technology.
ii
NASPI serves as a forum for information-sharing and problem-solving among the
synchrophasor projects and stakeholders. Much of the work and insights reflected in this
technical workshop was enabled by individuals and companies funded by DOEs on-
going research and development projects and the ARRA investments. Thus it is
appropriate to recognize the insights and work product documented in this workshop and
technical report as one of many consequences and work products resulting from the
federal Smart Grid investments. Therefore, the Department joins NASPI in re-releasing
this material to the smart grid community to document additional impacts and value
realized from the federal Smart Grid investments in synchrophasor technology.
iii
NASPI MODEL VALIDATION TECHNICAL WORKSHOP
WORKSHOP SUMMARY
With increased deployment of phasor measurement units (PMUs) across North Americas
bulk power system, utilities and grid operators are gaining new insights into grid and
asset behavior. This change results from PMU collection of high-speed, time-
synchronized data about grid conditions (voltage, current, frequency, and phase angles).
Model validation has been recently recognized as a highly successful use for
synchrophasor data, because model testing and improvement using actual grid
performance information is more accurate and often economical than traditional off-line
asset testing.
In October 2013, NASPI held a technical workshop to review the state of the art in model
validation, inviting leading practitioners and researchers to explain the model validation
process and share case studies in its use. Theses notes summarize those presentations,
which are attached to this summary for the readers review. The technical workshop was
webcast and the last three hours of the webcast have been archived here as a video
attachment on the Model Validation Workshop page.
Why do modeling?
Tom Burgess (NERC) opened the workshop. He explained that the power system is
designed and operated based on mathematical models that tell us the expected behavior
of power plants, grid elements, and the grid as a whole. When a generator or the system
does not act in the way that its model predicts, the mismatch between reality and model-
based expectations can cause severe disturbances and costly equipment damage.
Inaccurate models have contributed to a number of recent North American power
outages, including the 1996 WSCC outage (illustration below, Figure 1).
iv
Figure 1 -- WSCC August 1996 Outage -- actual event (top) and the simulation that
showed what planners expected would happen. (Source: BPA)
The time granularity and geographic specificity of synchrophasor data make it perfect for
model validation, allowing the analyst to benchmark and improve models against actual
system performance rather than hypothesized behavior. And better models improve
system security and asset utilization.
Bob Cummings (NERC) said that bad models lead to bad decisions in planning and
operations, and can lead us to operate the system -- unintentionally -- in an insecure state.
Unmodeled and therefore unanticipated generation behavior causes many grid
disturbances.
NERC has begun an initiative to improve and validate powerflow and dynamics models,
benchmarking them against actual system performance as measured by PMUs. NERC is
also working with industry to study the interaction of system protection and turbine
controls -- again, using actual system performance, measured by PMUs, for better
understanding. Because WECC has already made much progress in model validation,
much of this new work is focused in the electrically complex Eastern Interconnection,
where there is much to learn about governor and exciter models, load behavior, frequency
response, and inter-area oscillations. Another Eastern Interconnection priority should be
to improve the compatibility and ease of data exchange between regional system models
and asset models.
Vickie vanZandt (WECC) reports that we need better models because the current
electrical system is very complex and requires good models and simulations in order to
design appropriate operating limits and protection systems. Again, bad models foster
inappropriate system designs (Figure 2).
v
Figure 2 -- August 4, 2000 Oscillation that led to the separation of Alberta from the
rest of the Western grid -- the simulation led planners to expect that under
these conditions the oscillation would damp out, when in fact the oscillations
lasted much longer with violent result. (Source: WECC)
VanZandt also illustrated the application of synchrophasors for power plant model
calibration (Figure 3).
vi
Figure 3 -- Comparison of model fidelity and actual generator behavior for an 1,100
MW nuclear plant before and after PMU data-based model calibration
(Source: BPA)
Generator models
Dmitry Kosterev (BPA) reiterated that accurate power system models are required for
reliable, economic grid operation. NERCs MOD standards1, developed after the 2003
Northeast blackout, require formal model verification. An accurate generator model
requires both the correct model structure and an accurate set of data. Transmission
planners can use PMU recordings of system disturbances for independent verification of
models and dynamic performance.
Following the 1996 outages, WECC required that generator owners test their equipment
for the purpose of model verification. WECC differentiates between (i) baseline model
development and (ii) periodic model verification. For baseline model development, stage
testing is often required. Even then, as BPA experience has shown, staged tests can
produce inconsistent results, and PMU disturbance data is useful in supplementing the
model development. For periodic model verification, BPA finds PMU data to be a cost-
effective way to perform model verification, because the data are collected with the
generating unit in normal operation and do not affect its power production.2 In addition,
1
All of the current NERC reliability standards, including the MOD (modeling) standards can be
found at
http://www.nerc.com/pa/Stand/Reliability%20Standards%20Complete%20Set/RSCompleteSet.p
df.
2
Traditional generator testing has required the generator to be taken off-line, cutting power
production and revenues, and producing test results that can be inconsistent.
vii
model verification using PMU data can be done more frequently (up to 10 times per year
compared to once every 10 years under the current NERC Standards), maintaining high
overall system reliability.
WECC testing has revealed that plants with legacy analog controls have the most errors
and their settings and performance tend to change over time. In contrast, plants with
modern digital systems enter service with good models that stay accurate over time
(because the plants behavior changes less over time). Common sources of generator
model inaccuracy can include erroneous representations of power system stabilizers,
turbine control operations, governor models, and generator inertia. In some cases the
errors are due to deficiencies in the models structure (i.e., the characteristics of generator
behavior and the relationships between its elements have not been accurately
represented), and in other cases the model has been mis-calibrated (i.e., while the
structural elements are correct, some of the settings are inaccurate).
BPA has been using PMU data recordings of generator performance in response to grid
disturbances to validate dynamic models of power plant data. Insights from these events
complement baseline model development, and plant performance through multiple grid
events yield richer data that produces more accurate models. BPA has developed
software to automate model testing and validation using synchrophasor data.
Once a good power plant model is established, PMU recordings of system disturbances
can be used for clinical assessment of power plant performance and detection of
control failures, such as a failure of a power system stabilizer at a large hydropower
generator in Pacific Northwest (Figure 4). BPAs goal is to get a performance report
on its entire generating fleet within minutes of a system disturbance event.
Bob Zavadil (Enernex) has been leading work for the Utility Variable Integration Group
and the National Renewable Energy Laboratory to validate models of wind generators
viii
and photovoltaic plants; with the rapid growth in renewable generation capacity across
the continent, it is crucial that grid designers and operators can understand and predict
how these plants contribute to and respond to potential grid problems (particularly with
respect to local voltage). Oklahoma Gas & Electric has a significant amount of PMU-
recorded data on wind plant performance that is being used for this effort.
To validate a wind plant model, Zavadil starts with detailed information about the power
plant, and a generic model for the wind turbine. He aggregates the wind turbine model to
approximate the magnitude of the plant as a whole, and uses parameter sensitivity
analysis to iteratively adjust the wind turbine models until their modeled behavior
collectively resembles like the actual measured events.
The participation of plant owners and transmission operators is critical for model
validation, to get detailed information about both the plant and the grid. Zavadil points
out that although disturbance data is essential for model validation, there are not a lot of
events on the grid, so appropriate data for validation may be long in coming. For this and
other reasons, UVIG has begun a renewable plant model validation collaborative to share
model validation information and resources.
Pouyan Pourbeik (EPRI) and George Stefopoulos (NYPA) used generic models
developed by EPRI and NYPA to improve dynamic models for NYPAs static VAr
systems (the Marcy convertible static compensator and its SVC). Beginning with the
generic SVC models and PMU data from disturbance events, they calculated from the
PMU data the injected reactive current and reactive power of the SVC and chose the
device model accordingly. NYPA has also automated its model optimization process
with the Static VAr System Model Validation tool.
Xiaochuan Luo (ISO-NE) asserted that phasor data and on-line transient stability
assessment are the foundations for dynamic model validation. ISO-NE has used PMU
data to validate nuclear plant and HVDC models. Figure 5 shows the ISO-NE HVDC
model performance before and after validation, relative to actual PMU-recorded
performance during a single-phase ground fault. Similarly, they validated the Millstone
nuclear unit model using PMU data for a Phase B to ground fault that occurred 16 miles
away from the power plant.
ix
Figure 5 -- ISO-NE validation of HVDC model (Source: ISO-NE)
ISO-NE has also automated its model validation process using actual system event data,
and is now moving on to validate other models for generators, HVDCs, loads, and SVCs.
Bill Blevins (ERCOT) reports that ERCOT is using phasor data for model validation in
part because the PMUs record on-line events and activity that may not be observable in
off-line field generator testing. The ERCOT process goes beyond model validation, and
includes identifying voltage oscillations at the plant (including poorly damped
oscillations at low output and undamped oscillations at high output) and doing post-event
analysis to recreate the oscillations (through simulations). This allows them to identify
the causes and find solutions to mitigate the oscillations.
ERCOT is using this process to comply with NERCs MOD 26 and 27 requirements for
generator and exciter model validation, to assure that its dynamic models match actual
equipment in the field. ERCOT is tuning its models with parameter estimation and
verification. Blevins notes that reduction of its system models to data time-series
simplifies the model validation process and facilitates automated model validation.
Bernie LeSieutre (University of Wisconsin) has been working to refine the process of
model revalidation or invalidation and calibration using PMU disturbance data.
Comparison of the recorded actual event against the models predicted generator
performance allows the analyst to determine whether the models predictive capability is
so far off that it should be fully refuted, restructured, or recalibrated (Figure 6). He
showed examples of how to use PMU data to compare against the original model.
LeSieutre also showed results of validated models (with model fit improvements) that can
predict operational results that match actual historical performance (Figure 7). The
model fit process uses sensitivity models to understand model components and hone in
x
on whether the appropriate improvements might entail structural modification or
recalibration to modify the model parameters (Figure 8).
Figure 6 -- Comparing real data against the model -- black lines are original model
results, red lines are PMU data, blue lines are recalibrated model results (Source:
LeSieutre, University of Wisconsin)
Figure 7 -- Using PMU data to adjust model to the observations -- black lines are
original model results, red lines are PMU data, blue lines are recalibrated model
results (Source: LeSieutre, University of Wisconsin)
xi
Figure 8 -- Examples of parameter adjustments in model recalibration (Source:
LeSieutre, University of Wisconsin)
LeSieutre emphasized that models validated against data from multiple disturbances
produce better results. He said that even with an automated model validation processes,
engineering judgment is still essential to understand the models and know whether the
model needs calibration or structural improvement.
Dmitry Kosterev (BPA) explained that while a power plant model looks at only one
element on the grid, system models need to accurately represent not only individual grid
elements (generators, transformers, SVCs, etc.) and loads, but also how all those
elements will interact with each other. During a disturbance, the grid as a whole may act
in ways different from just the sum of its parts. System modeling is not yet as far
advanced as power plant modeling, in part because it is only recently that system
planners have viewed the relevant system scope as being an entire interconnection rather
than the footprint of a single transmission owner or reliability coordinator.
Bharat Bhargava, Kevin Chen and Anamitra Pal (all with Electric Power Group)
and Juan Castaneda and Farroukh Habibi-Ahsrafi (Southern California Edison)
have used data from three major western grid events to validate the dynamic system
response simulated by two phasor data applications: RTDMS and PGDA. Using the data
from the 2011 Pacific Southwest blackout, the January 2008 HVDC oscillations, and
simulations of the Pacific Intertie under stress, the team performed dynamic event
simulations using PSLF. They then streamed the data into the RTDMS visualization tool
and the PGDA off-line analysis tool. They report that these simulations can be used
xii
effectively to compare and validate models and to feed event replays for operator
training.
Dmitry Kosterev (BPA) and his team have been using PMU data on past oscillations
and Fault-Induced Delayed Voltage Recovery events to improve the load model in GEs
PSLF tool and Siemens PSS/E. They caution that while we can tune and improve load
models, load behavior is so complex that we cant fully predict how load will behave.
Additionally, since loads are connected to individual electrical phases on the system, we
need to collect and analyze point-on-wave data rather than positive sequence data to
perform accurate load analysis. BPA and WECC are working with the DOE CERTS
program and SCE to collect such data using distribution-level power quality monitors.
Eric Allen (NERC) warns that without periodic updating and evaluation, models that
started out relatively accurate can drift from actual system behavior over time through
the collective impacts of new load dynamics and generator dynamics. Over time, a few
insignificant discrepancies in the models of individual grid elements can accumulate
and, in aggregate, produce significant errors in system model estimates.
Planners today are increasingly using dynamic disturbance data to test system dynamics
models. To do this they develop a powerflow case to represent system conditions
preceding the disturbance and perform dynamic simulations with those starting
conditions. They then iterate those simulations to adjust them against actual events using
parametric analysis and adjustments. The planners goal is to produce a model with high
fidelity relative to actual events.
xiii
NASPI TECHNICAL WORKSHOP
8:30 am to 12:30 pm
OHare V Ballroom
This workshop will provide a detailed grounding in the benefits of using synchrophasor
data for electric system and power plant model validation and explain the process and
steps for doing so. Presenters will address recent requirements and opportunities for
model validation, provide an overview of generator and power system models and model
validation tools, and explain what synchrophasor data are needed for power plant and
system model validation. Several speakers will provide detailed briefings on the process
and results of several specific cases where synchrophasor data have been successfully
used for power plant model validation, identification of inappropriate asset operations,
and dynamic grid models.
If you wish to attend this technical workshop, please register at model val workshop;
there is no registration fee.
If you cannot join us in person for this workshop, you can follow the presentations in real
time through webinar access -- use Join Lync Meeting
(https://lcmeet.pnnl.gov/teresa.carlon/RYJ8RKQM). Remote participants will not be
able to interact with the presenter. The workshop presentations will be posted on the
NASPI website (www.naspi.org) and we will attempt to archive and post the webinar as
well.
The Work Group meeting of the North American SynchroPhasor Initiative, which will
feature progress reports from the North American synchrophasor project grant recipients,
technical sessions and a vendor trade show, will begin on the afternoon of October 22
(following this workshop) and run through noon on October 24 in the Crowne Plaza
Chicago OHare Hotel. The NASPI Work Group meeting will require separate
registration (WG meeting reg) and a fee of $350 for late registrants.
1
!
NASPI MODEL VALIDATION TECHNICAL WORKSHOP AGENDA
8:40 am Intro to power plant models and grid models -- Bob Cummings (NERC)
8:55 am Why use synchrophasor data for model validation -- Vickie vanZandt
(WECC)
9:05 am Expectations and practicalities for using phasor data -- Dmitry Kosterev
(BPA)
9:15 am The basics of plant model validation using PMU disturbance data
Dmitry Kosterev (BPA) Value of using PMUs for model validation and
detection of control abnormalities; data required for model validation and
calibration; steps required to set up model validation and generator
performance monitoring process by Transmission Planner and Generator
Owner; WECC and BPA case studies.
10:00 am Break
10:15 am Case study 1 -- Wind power plant model validation -- Bob Zavadil (for
UVIG, using OG&E wind plant data) -- current model validation efforts
(scope, what kinds of wind plants and turbines being studied), how
applicable plant-specific data and model results are to other wind plants,
why it's needed, what data they're using, how it's going, when it'll be done,
what's next.
11:05 am Case study 4 -- ERCOT using phasor data to find inaccuracies in generator
models -- Bill Blevins (ERCOT)
11:30 am Case study 5 using phasor data for power plant model calibration and
PSS failure detection -- Bernie Lesieutre (University of Wisconsin)
2
!
11:45 am Case study 6 -- using phasor data and simulations in the RTDMS and
PGDA programs to validate system response and dynamic models --
Bharat Bhargava (EPG)
12:00 pm Power system dyamic model validation -- Eric Allen (NERC) and Dmitry
Kosterev (BPA) -- what model used, why it needs validation, what it
takes to develop a validation base case, what synchrophasor data being
used, what's the process for doing this, how long will it take to get a model
you're happy with, how much of the calibration process requires getting
the underlying grid components modeled accurately rather than working
on the synergistic results?
12:30 pm Adjourn
3
!
The Importance of Modeling
Thomas Burgess
Vice President and Director, Reliability Assessment and Performance
Analysis
NASPI Model Validation Workshop
October 22, 2013
Importance of Modeling
Real event
Dynamic
simulations
4000
2500
MW
(
2000
1500
1000
500
0
15:27:30.240 15:28:56.640 15:30:23.040 15:31:49.440 15:33:15.840 15:34:42.240 15:36:08.640 15:37:35.040
Local Time (PDT)
1. Generator Dynamics
2. *Load Behavior
3. *Frequency Response
4. *Inter-Area Oscillations
5. *Equipment Modeling
6. *Special Protection Systems/Remedial Action
Schemes
7. *Protection Systems
8. *Turbine and Boiler Controls
Components of a Broad Modeling Initiative Design
)
5 RELIABILITY | ACCOUNTABILITY
Frequency Response Modeling
Actual
Models Adjusted
Database Models
6 RELIABILITY | ACCOUNTABILITY
Role of Synchrophasors in
Enhanced Modeling
10 RELIABILITY | ACCOUNTABILITY
Operational Modeling and Model
Inputs
Robert W. Cummings
Director, Reliability Initiatives and System Analysis
NASPI Model Validation Workshop
July 11, 2013
Importance of Modeling
2 RELIABILITY | ACCOUNTABILITY
August 10, 1996 WSCC Outage
3
Real event
Dynamic
simulations
3
No confidence in dynamic database
RELIABILITY | ACCOUNTABILITY
WSCC Actions Since 1996
4
Aggressive testing of generating units
80% of units directly tested
Validation by Observation adopted
System probing testing
Pacific DC Tie (PDCI) signal injection (ongoing process)
Chief Joseph Braking Resistor (1,400 MW) insertion
60.1
- COI power
60
COI Power, June 14 2004 West Wing Disturbance
6000
59.9
5500
Frequency [Hz]
Actual
59.8
Malin Freq 5000
4500
59.7
Power [MW]
Actual
4000
COI
59.6
3500
59.5
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 3000
Time [sec]
2500
Real event
2000
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75
5 RELIABILITY | ACCOUNTABILITY
Unable to Simulate EI Frequency Response
6
Highlighted in December 2011 FERC report
Highlighted
simulationin recentsignificantly
predicted FERC report greater frequency
simulation
response than predicted
was,significantly greater frequency
in fact, recorded response than
by monitoring
was, in fact, recorded by monitoring equipment.
equipment.
6 RELIABILITY | ACCOUNTABILITY
MRO Disturbance Sept. 18, 2007
Actual (DFR)
20% Governor
Response
As-modeled
Governor
Response
7 RELIABILITY | ACCOUNTABILITY
Actual Aug. 4, 2007 Frequency
60.03
60.01
59.99
59.97
59.95
Hz
59.93
59.91
59.89
59.87
59.85
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64
Seconds
8 RELIABILITY | ACCOUNTABILITY
Governor Modeling
Base
Base
(Legacy
(Legacy
Models)
Models)
Best
Best
Generic
Generic
Model
Model
9 RELIABILITY | ACCOUNTABILITY
Un-modeled Generation Behavior
Shakespearean generation
How can I trip thee, let me count the ways
10 RELIABILITY | ACCOUNTABILITY
Power-Load Unbalance Control Function
1.000
59.9
0.800
59.8
Frequency
Per Unit
0.600
59.7
0.400
59.6
0.200 Unit 2 MVA decreases at a rate
of 3.37pu/sec
freq(min) = 59.642, Max Unit MW, 59.5
0.000 and Min Unit MVAR @
16:44:16.3667 at the same time as
Rockport-Sullivan trip
-0.200 59.4
25 7
25 2
25 7
25 4
25 7
25 6
25 7
25 8
25 7
25 7
2
25 6
25 7
06
13
26
33
46
53
66
73
86
93
06
13
6.
6.
6.
6.
7.
25
25
25
25
25
25
25
6.
6.
6.
6.
6.
6.
6.
6.
6.
6.
7.
7.
time in seconds from 16:40:00.000
11 RELIABILITY | ACCOUNTABILITY
Modeling Initiative
14 RELIABILITY | ACCOUNTABILITY
Modeling Gap Analysis
15 RELIABILITY | ACCOUNTABILITY
16 RELIABILITY | ACCOUNTABILITY
The Reasons We Need Better Models
2
Premise This Isnt Your Mothers
Power System Anymore
The Complex New Days:
o Smaller, more distributed generation for which the
grid was not designed
o Many more transactions that change in increments
of 5-10 minutes
o The generation fleets characteristics have changed
a greater percentage of intermittent, low mass
machines less inertial response to help arrest
frequency decline
o Finally, the load has changed less industrial,
voltage dependent load, and more
3 computer and air conditioning service
So What Does That Mean?
A grid that is more complex and harder to
operateand demands better modeling.
No matter how carefully operators,
operating engineers, and planning
engineers study the system.if the
models arent right,
.the results they get and the limits they
set arent right either.
4
So What Does That Mean?
Of the three components,
o Transmission
o Generation
o Loads
Transmission is pretty good (status of
MODs notwithstanding)
Generation is improving, but more to go
Loads need the most work
5
So What Does That Mean?
SCADA cant help much with this effort
More frequent and time-synchronized
measurements are necessary to get this
model improvement done
We happen to have some of those coming
in..
6
7
Happen 3300
(Simulation) 3200
3100
20 30 40 50 60 70 80 90 100 110 120
Time [sec]
3700
This Is What Power on AC Intertie at California-Oregon
Actually
3600
Border - Actual Oscillation Measured
Happened 3500
Power [MW]
August 4, 2000
3400
Oscillation -
Alberta Separation 3300
3200
3100
20 30 40 50 60 70 80 90 100 110 120
Time [sec]
9
.and now, on with the Case Studies
10
PowerPlantModelValidation
andPerformanceMonitoring
DmitryKosterev,SteveYang,Pavel Etingov
NASPIWorkshop
October2103
PowerSystemModels
Accuratepowersystemmodelsarerequiredfor
0 Failureofmodelstopredict August10,1996
4400
0 7.4Mcustomerslostpower 4200
duetotheoutage 4000
ratedtemporarilyby33% 4400
0 WSCCBOTrequiredall 4200
4000
generators>20MVAbe 0 10 20 30 40 50 60 70 80 90
Requirements
1996 generatorbaselinetestingformodelverificationis
requiredinWSCC
BenefitsofWECCgeneratortestingprogramare
indisputable:
Vastmajorityofmodelsneededrevisions
Structuralmodelerrorsweredetected
Errorsincontrolsettingswereidentifiedandcorrected
Needtosustainthemodelvalidationwasapparent
2006 WECCformalizeditsGeneratingUnitModel
ValidationPolicy
BaselineModelDevelopment
PeriodicModelValidation
ReliabilityStandards
2007 NERCstartedthedevelopmentofmodel
verificationstandards
2013 NERCapproved
NERCMOD025 reactivepowercapabilitiesverification
NERCMOD026 generatorandexcitationcontrolmodel
verification
NERCMOD027 generatorturbinecontrolmodel
verification
NERCPRC019 coordinationofgeneratorprotectionand
controls
2013 NERCMODBefforttoaddressFERC
directives
Requiresplantoperatortoprovideaccuratemodeldata
Perspectives
GeneratorOwner/Operator
Ownsandoperatesgeneratingunit
Hasknowledgeoftheirgeneratingequipment
Responsibletoprovideaccuratemodelstoatransmission
planner
TransmissionPlanner
Usesmodelsinsystemstudies
Needstoverifythatthemodelsareusable
Maywanttohaveanindependentwaytoverifymodel
accuracy
GeneratorOwner
GeneratorOwner:
BaselineTestingvs.ModelValidation
Baselinemodeldevelopment
Neededtoestablishthecorrectmodelstructure
Neededtocreateinitialmodeldataset
Periodicmodelvalidation
Donetoensurethatthemodelsstayaccurateandup
todateAFTERagoodmodelbaselineisdeveloped
Shouldnotbeasubstituteforbaselinemodel
development
BaselineModelDevelopment
Equipment Model
Speed Reference
Z 1 1 1
6 1 + s Tf
6 Tp s
ucl Gmin
Damping PILOT
SERVOMOTOR
s Tr
Rt 1 + s Tr
Rp
Neededtoestablishthecorrectmodelstructure
Inspectionofequipmentandcontrolsettings
Sometestsarerequired
Disturbancemonitoringcancomplementmodeldevelopment
8
EASY
NOT
GeneratorOwner
Werecommendgeneratorownerstorequiretest
andrecordingcapabilitiesinnewdigitalexcitation
systemsandgovernors
Needtoensurerecordinghasadequatebandwidth
Westronglyencouragegeneratorownerstoinstall
disturbancemonitorsinapowerplant
Statorthreephasevoltagesandcurrents
Fieldvoltageandcurrent
Governorvalveposition
Contact:ShawnPatterson,USBR
TransmissionPlanner
Thesamepowerplanttestedbytwoconsultants
Active Power
170
/
165
/
Power (MW)
160
/
155
/ ConsultantA
150
/
10 15 20 25 30 35 40 45 50 55 60
/
-60
Reactive Power
ConsultantB
-80
Power (MW)
-100
-120
-140
10 15 20 25 30 35 40 45 50 55 60
/
Time (sec)
/
Whichdataiscorrect?
Youdonotknowunlessyouhaveanindependentwayofverifying
UsingPMUDataforModelValidation
BPAhasinstalledPMUsatpowerplantPOIs
BPAdevelopedPowerPlantModelValidation
(PPMV)applicationusingPMUdata
Record:
POIbusvoltage
POIbusfrequency
Pointof PowerplantMWsandMVARs
Interconnectio V I
n
~
Turnedoutneitherconsultantwasright
Active Power
170
/
165
/
Power (MW)
160
/
155
/ ConsultantA
150
/
10 15 20 25 30 35 40 45 50 55 60
/
-60
Reactive Power
ConsultantB
-80
Power (MW)
-100
-120
Reality
-140
10 15 20 25 30 35 40 45 50 55 60
/
Time (sec)
PowerPlantModelValidation
Whatagoodmodelslookslike:
Voltageandfrequencyareinputs
Activeandreactivepoweraremeasuresofsuccess
Blueline=actualrecording
Redline=model
PowerPlantModelValidation
Whatabadmodellookslike:
Voltageandfrequencyareinputs
Activeandreactivepoweraremeasuresofsuccess
Blueline=actualrecording
Redline=model
BPAExperiencewithDisturbanceBased
ModelValidation
Mostcommonmodelissues:
PowerSystemStabilizermodels
Turbinecontrolmodeofoperation/governormodels
Generatorinertia
Deficienciesinmodelstructure
Otherreasonsformodelmismatch
AutomaticGenerationControls
UEL
Clinical experience:
Plantswithmoderndigitalsystemshavegoodmodels
thatstayaccurateovertime
Plantswithlegacyanalogcontrolshavemosterrors
andtendtochangeintime 18
FrequencyResponsivePlant
Plant
Frequency
4
60.1
4
Frequency (Hz)
60
4
59.9
59.8
59.7
0 10 20 30 40 50 60 70 80 90 100
4
Power
4
340
4
Power (MW)
330
4
320
4
310
4
0 10 20 30 40 50 60 70 80 90 100
4
Time (sec)
Unit
Frequency
4
60.1
4
Frequency (Hz)
60
4
59.9
59.8
0 10 20 30 40 50 60 70 80 90 100
4
Power
4
800
4
Power (MW)
750
4
700
4
650
4
0 10 20 30 40 50 60 70 80 90 100
4
Time (sec)
Control
Frequency
4
60.1
Frequency (Hz)
60
4
59.9
59.8
0 10 20 30 40 50 60 70 80 90 100
4
Power
4
420
4
400
4
Power (MW)
380
4
360
4
340
4
0 10 20 30 40 50 60 70 80 90 100
4
Time (sec)
PerformanceMonitoringandDetecting
GeneratorControlFailures
Onceagoodbaselineisdeveloped,PMUisusedfor
clinical assessmentofpowerplantperformance
Controllerstatusatthe
generatorwasindicating
normalstate
PMUdisturbancedata
indicatedactualresponsevery
differentfromwhatwas
expected
Powerplantwascontacted,
Blueline=actualresponse controlsinspected,found
Redline=expectedresponse internalfailure
PerformanceMonitoring
Event1
PSS ON
PSS OFF
ActualPMUrecording,SimulationwithPSSON,SimulationwithPSSOFF
PerformanceMonitoring
Event3
PSS ON
PSS OFF
PerformanceMonitoring
Event4
PSS ON
PSS OFF
PSSfailedsometimebetweenevent3andevent4
PerformanceMonitoring
Event7
PSS ON
PSS OFF
BenefitsofPMUbasedModelValidation
Disturbancerecordingscancomplementthebaselinemodel
development(e.g.TransAlta BPAworkatCentralia)
PMUbasedmodelvalidationisanacceptablemethodfor
GOstocomplywithNERCMOD026,027
assumingacorrectbaselinemodelisdeveloped
PMUbasedmodelvalidationcanbeusedbyTPsto
independentlyverifythatthemodelsprovidedbyGOsare
accurate
BPAexperiencesuggeststhat60to70%ofmodelsdidnotmatch
disturbancerecordingsevenafterthebaselinetestwasperformed
TPsneedindependentmethodofmodelverification itisdifficult
topolicetrafficifyoudonothaveaspeedradar
PMUbasedmodelvalidationallowsmorefrequentmodel
verificationanddetectionofcontrolfailures(e.g.Grand
CouleeandColstrip)thanonceevery10years(perNERC)or5
years(perWECC)
28
WindPowerPlants
WindPowerPlantModelValidation
BPA,IdahoPowerinstalledseveralPMUsatwind
powerplants
BPAiscollaboratingwithEPRI,NREL,Enernex,UVIG,
Sandiaonwindpowerplantmodelvalidation
Initialresultssuggestmoremodeldevelopment
workisneededbeforemodelscanbeusedin
dynamicsimulations
WindPowerPlantModelValidation
Demonstration
Demonstration
PPMVa_SetBaseCase_v1a.p
PowerflowTemplate:
PLANT.sav
PSLF
SCADADataFile: Mini PowerflowFile:
SCADA_PLANT_EVENT.dat
State PLANT_EVENT.sav
Estimate
EventDataFile:
PMU_PLANT_EVENT.csv
Setupapowerflowwithinitialconditions
PMUDataFile
5
TimeVactFactPactQact//Head
1500 60 11//Scale
00000//Offset
00000//Tf
00.80.990200//min
1601.21.011000200//max
11111//Plot
0,542.696899,59.987999,561.183899,38.693913
0.033333,542.686523,59.988998,561.175293,38.754639
..
SCADADataFile
BusNumber
BusName
BaseKV
UnitID Baseloaded
UnitStatus or
UnitMW Responsive
UnitMVAR
102"GT01"18.00"1"1 165.04.2B
103"GT02"18.00"1"1 155.74.9B
104"ST12"16.00"1"1 236.28.5B
PPMVa_RunValidation_v1a.p
PowerflowFile:
PLANT_EVENT.sav
PlantDynamicData:
ChannelFile:
PLANT.dyd
PSLF PLANT_EVENT.chf
EventDataFile:
PMU_PLANT_EVENT.csv
Runpowerplantmodelvalidation
ModelCalibration
ModelCalibration
Initially,BPAuseofthePMUdatahasbeenlimited
tovalidatingdynamicmodelsofpowerplants:
usedforpass/failchecking
nomodeladjustmentsaremadeshouldthemodelbe
wrong
Model(in)Validation
Simulationsdoneusingamodelfrom
WECCdynamicdatabase
Blue=actual
Red=simulated
ModelCalibration
DOEisfundingseveralresearcherstodoworkon
powerplantmodelcalibrationusingPMUdata
PNNL(Kalmanfilter)
SakisMeliopolis,GeorgiaTech(supercalibrator)
BernardLesieutre,UniversityofWisconsin(pattern
matching)
WeiJenLee,UniversityofTexas(particleswarm
optimizationandnonlinearoptimization)
EPRIisalsoworkingonPMUbasedmodelcalibration
BPAhasworkedwithBernieLesieutretoperform
modelcalibrationforCGSandColstrip
ModelCalibration
Simulationsdoneusingacalibrated
model
Blue=actual
Red=simulated
ModelCalibration
Simulationsdoneusingacalibrated
model
Blue=actual
Red=simulated
ContactInformation
DmitryKosterev,BPA,dnkosterev@bpa.gov
SteveYang,BPA,hyang@bpa.gov
Pavel Etingov,PNNL,Pavel.Etingov@pnnl.gov
BernieLesieutre,UniversityofWisonsin
ShawnPatterson,USBR
NASPI TECHNICAL WORKSHOP:
*
MODEL VALIDATION USING
SYNCHROPHASOR DATA
&
TUESDAY, OCTOBER 22, 2013
*
Crowne Plaza Chicago OHare Hotel
*
5440 N. River Rd.
*
Rosemont, Illinois 60018
*
Robert M. Zavadil
Vice-President & Principal Consultant
bobz@enernex.com
www.enernex.com
3
2013 EnerNex. All Rights Reserved. www.enernex.com
OG&E PMU Data
f Large number of Complex event record with embedded large disturbance
recorded events
screened
f Many were small
signal i.e. slight
changes in terminal
voltage
f Looking for large
disturbances
4
2013 EnerNex. All Rights Reserved. www.enernex.com
Analysis
fBasic information about
plant obtained from
OG&E
fType III generic model
used to represent
turbines Measurement
fParameter sensitivity
analysis conducted to
iteratively adjust
aggregate turbine model Simulation
5
2013 EnerNex. All Rights Reserved. www.enernex.com
Results
fSimulation/measurement correspondence is
reasonable, but
fMaybe more of a supporting data point than
validation
fWhat is validation, anyway?
6
2013 EnerNex. All Rights Reserved. www.enernex.com
Solar Plant Validation
using Generic PV Plant Model
Real Power Reactive
7
2013 EnerNex. All Rights Reserved. www.enernex.com
Lessons Learned
f Even with wide-scale deployment of PMUs, good data for
validation is hard to come by.
f Good data is important, but not the only information
requirement
f Participation of Transmission owner, plant operators in
validation process would be very beneficial
f 1st generation of generic models may be lacking (good
news: 2nd generation imminent)
f Validation process itself needs more formalization
8
2013 EnerNex. All Rights Reserved. www.enernex.com
Challenges
f A specific event may be hard to replicate via simulation
Plant model complexities
Initial conditions/system state
Origin and nature of system disturbance
f Actual events will be asymmetrical
PSS/E, PSLF models are positive sequence only
Unbalanced events model very approximately
3-phase faults are rare
f Events are infrequent
With just a few monitored locations, appropriate data for validate may be long
in coming
Can be partially remedied by monitoring at many locations
f Large number of commercial turbines to validate (60 GW + wind, 10 GW
solar installed capacity = 100s of bulk power plants)
www.variablegen.org
11
2013 EnerNex. All Rights Reserved. www.enernex.com
Approaches for Model Validation
Developed thorough collaboration with WECC and vendors [1] & [2]
Calculating P, Q, I, and B
Vt = Vt *
S = 3 Vt I t
It = It
P = real(S)
Q = imag(S)
V t = Vt Vnom
ISVS = Q Vt
2
BSVS = Q Vt
PMU
Losses
(neglected
in stability
Calculated
simulations)
Calculate from the PMU data the injected reactive current (or
susceptance for SVC) and reactive power of SVS
Play the measured voltage back into the model and fit the simulated
reactive current I (or susceptance B) and Q to the measured values
Optimize the gains of the controllers to get a good match via least
squares estimation
[2] Generic Static Var System Models for the Western Electricity Coordinating Council,
April 2011. http://www.wecc.biz/committees/StandingCommittees/PCC/TSS/MVWG/
SVCTF/Shared%20Documents/GenericStaticVarSystemModelsforWECC.pdf
[3] Static Var System Model Validation (SVSMV) Version 2.0, June 2013, EPRI Product
ID 3002001009. http://www.epri.com/abstracts/Pages/ProductAbstract.aspx?
ProductId=000000003002001009
Xiaochuan Luo
TECHNICAL MANAGER
Outline
2
ISO-NEs Synchrophasor Infrastructure and Data
Utilization (SIDU)
ISO-NE PDC
External Entity PhasorPoint
ISO-NE
Network
DQMS ROSE
3
Pilot On-Line Transient Stability Assessment
Fixed data (updated periodically)
Network Dynamic
Generator
model in data in
mapping
planning planning
table
case case
Dynamic
EMS Snapshot Data Preparation Tool (DPT) equivalent for
external area
4
Validation of HVDC Model
Single phase-to-ground fault occurred at 18.2% of line 337 from
SANDY POND station
Fault was cleared after 5 cycles by opening line 337
5
Validation of HVDC Model
Sandy Pond 345 kV Voltage - PMU
6
Validation of HVDC Model
100
PMU measurement
0 Simluation with original HVDC model
Simulation with modified HVDC model
-100
Branch Active Power Flow [MW]
-200
-300
-400
-500
-600
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time [s]
7
Validation of Nuclear Unit
Phase B to ground fault on line 348, 16 miles away from the MILLSTONE station
Fault was cleared after 5 cycles by opening line 348; Line was reclosed after 10.5
seconds
8
Millstone 345 kV Voltage
1.05
PMU
Simulation
1
Voltage
0.95
0.9
0.85
30 32 34 36 38 40 42 44 46 48
Time [s]
9
Line 310 Active Power (Millstone Manchester)
650
Line 310 PMU
Line 310 Simulation
600
550
Active Power Flow [MW]
500
450
400
350
300
30 32 34 36 38 40 42 44 46 48
Time [s]
10
Line 368 Active Power (Manchester Card)
-200
Line 368 PMU
Line 368 Simulation
-250
-300
Active Power Flow [MW]
-350
-400
-450
-500
30 32 34 36 38 40 42 44 46 48
Time [s]
11
Millstone Unit 3 Active Power
-1050
U3 PMU
U3 Simulation
-1100
-1150
Active Power Flow [MW]
-1200
-1250
-1300
-1350
30 32 34 36 38 40 42 44 46 48
Time [s]
12
Millstone Unit 2 Active Power
-760
U2 PMU
-780
U2 Simulation
-800
-820
Active Power Flow [MW]
-840
-860
-880
-900
-920
-940
30 32 34 36 38 40 42 44 46 48
Time [s]
13
Conclusion and Future Plans
14
15
Use of Synchronized Phasor Measurement for
Model Validation
Bill Blevins
Manager
ERCOT Operations Planning
NASPI Meeting
October 22 2013
Overview of the ERCOT System
Post-event analysis
Real time system monitoring
Generator model validation
Model Re-Validation/Invalidation
and Calibration using PMU
Disturbance Data
Bernie Lesieutre
Affirm/Refute/Calibrate
Compare Data and Simulations
Refute Model
Recalibrate Model
| 4
Recalibration
Sensitivity Models
We rely on models to analyze the grid under
various conditions - actual and anticipatory.
Model Tuning
Weve calculated sensitivity models to help
6
| 7
Before
After
7
| 8
Kqs
0 0.3 0.6 0.9 1.2 1.5 1.8
Tq
0 5 10 15 20 25 30
Tq1
0 0.03 0.06 0.09 0.12 0.15 0.18
8
| 9
Event Voltages
| 11
Event Frequencies
| 12
Event 1, part 1
Original
Parameters
Modified
Parameters
| 13
Event 1, part 2
Original
Parameters
Modified
Parameters
| 14
Event 1, part 2
| 15
Event 3
| 16
Event 4
| 17
Event 5
Original
Parameters
Modified
Parameters
| 18
Validation: Event 2
| 19
Validation: Event 6
| 20
Second Plant
Data from eight events over 14 months
Parameters estimated using first two events
Match is poor for events 4-8.
Match is good when PSS gain set to zero for
events 4-8.
| 21
Event 1
No PSS
| 22
Event 2
No PSS
| 23
Event 3
No PSS
| 24
Event 4
No PSS
| 25
Event 5
No PSS
| 26
Event 6
No PSS
| 27
Event 7
No PSS
| 28
Event 8
No PSS
| 29
Example 2 Conclusion
For this plant, the simulations are consistent
with the hypothesis that the PSS is operating
differently for the latter events than for the
earlier events.
The process also involved recalibration of
parameter values.
| 30
Event 1
Original
Parameters
Modified
Parameters
| 34
Event 2
Original
Parameters
Modified
Parameters
| 35
Event 3
Original
Parameters
Modified
Parameters
| 36
Event 4
Original
Parameters
Modified
Parameters
| 37
Event 5
Original
Parameters
Modified
Parameters
| 38
Event 6
Original
Parameters
Modified
Parameters
| 39
Event 7
Original
Parameters
Modified
Parameters
| 40
H 4.42 4.42
Xc 0.052 0.052
Ka 443 400
Kf 0.021 0.021
Tf 0.79 0.78
T5 12.6 12.6
Ks 8.23 8.23
Rp 0.045 0.045
Rt 0.46 0.50
Tr 1.82 1.65
Tw 1.76 1.76
| 43
H 4.42 4.38
Xc 0.052 0.052
Ka 443 400
Kf 0.021 0.021
Tf 0.79 0.77
T5 12.6 12.6
Ks 8.23 8.25
Rp 0.045 0.045
Rt 0.46 0.52
Tr 1.82 1.67
Tw 1.76 1.27
| 44
H 4.42 3.62
Xc 0.052 0.052
Ka 443 400
Kf 0.021 0.020
Tf 0.79 0.71
T5 12.6 12.6
Ks 8.23 8.27
Rp 0.045 0.045
Rt 0.46 0.50
Tr 1.82 1.74
Tw 1.76 1.27
| 45
H 4.42 4.47
Xc 0.052 0.052
Ka 443 400
Kf 0.021 0.021
Tf 0.79 0.72
T5 12.6 12.6
Ks 8.23 8.26
Rp 0.045 0.045
Rt 0.46 0.40
Tr 1.82 2.02
Tw 1.76 2.00
| 46
Training Signal 1
H 4.42 4.47
Xc 0.052 0.046
Ka 443 500
Kf 0.021 0.027
Tf 0.79 0.89
T5 12.6 12.2
Ks 8.23 8.58
Rp 0.045 0.045
Rt 0.46 0.32
Tr 1.82 2.61
Tw 1.76 2.04
| 48
Conclusions
PMU data is absolutely valuable for
affirming, refuting and recalibrating
models.
Engineering judgment is (still) required to
understand models, to
know which knobs to turn in calibration, and how
far,
Identify structural changes in the data.
Using Phasor System Data in RTDMS & PGDA
to Validate System Response and
Dynamic Models
Bharat Bhargava, Kevin Chen, Anamitra Pal
In Collaboration With
Juan Castaneda
Farroukh Habibi-Ashrafi
Page 1
Oct 2013
RTDMS and PGDA Overview for Simulations
Real Time Dynamic Monitoring System (RTDMS) is used for visualization and Phasor
Grid Dynamics Analyzer (PGDA) is used for detailed off-line analysis
RTDMS typically takes C37.118 / 61850 high-speed synchro-phasor system data and
can display multiple parameters important for operation of the power system
PGDA takes multiple formats such as dst, comtrade, csv, synchro-phasor system data
and supports detailed analysis of the system event
For simulations, EPG developed capability to import csv file formats in both these
programs
These programs can now be used to visualize and analyze PSLF simulations by
converting PSLF output data in to CSV files
Simulated data can be compared with actual events and can be used for system
model validation
Extreme events can be simulated and run using RTDMS to train operators
Page 2
Oct 2013
Overview of Methodology to Validate System Response and
Dynamic Models Using RTDMS and PGDA
Page 3
Oct 2013
Methodology Parameters Used in Visualization
Basic
Voltage Magnitude
Voltage Angle
Frequency, df/dt
Power
Reactive Power
Advanced
Modes of Oscillations and their Damping
Voltage sensitivity (dV/dP100 kV change per 100 MW)
Angle sensitivity (dA/dP100- degrees change per 100 MW)
Page 4
Oct 2013
Methodology Parameters Used in Off-Line Analysis
Basic
Voltage Magnitude
Voltage Angle
Frequency Transients
Frequency Response
Real Power
Reactive Power
Advanced
Oscillation and Damping
Mode Meter Ambient Oscillation Analysis
Ring Down Analysis
Spectral Analysis
Page 5
Oct 2013
Methodology Use Cases
Page 6
Oct 2013
September 8, 2011 Pacific Southwest
Blackout Event Simulation
Event Description:
Event took about 12-minutes
Complete event simulated using PSLF
Compared with the NERC/FERC report
Simulation matches very closely
Simulation replayed in RTDMS and analyzed using PGDA for
validation
EPG is working with SCE to simulate the event using their Real
Time Digital Simulator (RTDS)
Sequence of events in simulation includes:
Outage of North Gila-Hassayampa line
Outage of IID transformers
Load drop in IID and CFE
Loss of CFE and IID generation
Separation of SDGE at San Onofre Power Plant
Page 7
Oct 2013
Comparison of Actual (NERC/FERC Report)
and Simulated September 8, 2011 Event
8700 amperes
8400 amperes
Drop of 444 MW of
load in IID area
1 2 3 4 5 6 7
Page 8
Oct 2013
Visualization of San Diego Blackout Simulation
(After Hassayampa-N.Gila Line Trip) - Replay
About 70 MW peak to
peak oscillations on
each Malin-Round
Mountain lines
Damping below 2
percent
Frequency
Mode
Amplitude
Frequency oscillations
occurring at Colstrip
Frequency
Governor /Mode
Machine controls
Frequency
causing 50+ MW oscillations on
two generators
Mode
Damping
Mode
Amplitude
As seen in the above table, the Voltage at Malin substation sags as the COI is
loaded and Angle difference between Grand Coulee and Devers increases
Page 16
Oct 2013
Power and Voltage Plot at Malin Substation
Low damping of
0.6 Hz mode
Oscillations detected at
0.28 Hz
Malin 5680 MW
4860 MW
COI
Devers
Grand Coulee, WA
Low damping of
0.6 Hz mode
Low damping of
0.6 Hz mode
Page 22
Oct 2013
Thank You!
For questions, please contact Bharat Bhargava
Bhargava@ElectricPowerGroup.com
October 2013
Composite Load Model (CMPLDW)
M
12.5-kV
13.8-kV M
69-kV
115-kV
M
138-kV
AC
UVLS
UFLS Electronic
GE PSLF
Siemens PTI PSSE
Power World Static
PowerTech TSAT
Where we are now
WECC Composite load model version 1 is implemented in GE
PSLF and Siemens PTI PSSE, similar models exist in Power
World, Power Tech TSAT
Improvements to LTC models are requested in GE PSLF
Minor modifications are suggested for PSSE Model
SCV
SCC
DSW
4
WECC Load Composition Model
5
Load Model Validation
Studies
August 4, 2000 Oscillation Old Model
545
543
541
539
537
Voltage (kV)
535
533
531
529
527
525
0 5 10 15 20 25 30 35 40
Time (sec)
544
542
540
538
536
534
532
530
528
526
524
0 5 10 15 20 25 30 35 40
10
Data for Load Model Validation
Positive sequence data is no longer sufficient